Stochastic Calculus and Black-Scholes Theory MTH772P

Solutions Exercises Sheet 3

1. A k-dimensional Brownian motion is a random vector-function (B1 (t), . . . , Bk (t)),
where Bi are k independent (standard) BM. For smooth function f of the k-dimensional
BM the stochastic differential is
df (B1 (t), . . . , Bk (t)) =

k
X
i=1

1X
fx ,x (B1 (t), . . . , Bk (t))dt.
fxi (B1 (t), . . . , Bk (t))dBi (t)+
2 i=1 i i

(This follows by taking the Taylor expansion up to the second order terms and using
the rules (dBi )2 = dt, dBi dBj = 0 for i 6= j).
qP
k
2
The Bessel process with parameter k is defined as X(t) =
i=1 Bi (t), which is
the radial part of the k-dimensional BM.
(a) Show that X satisfies the stochastic differential equation
dX(t) =

k
X
Bi (t)

X(t)

i=1

dBi (t) +

n1
dt
X(t)

(b) Show that X satisfies the stochastic differential equation with single BM B
dX(t) = dB(t) +

n1
dt.
X(t)

Hint: use Levys theorem which states that a continuous martingale M (t) with
quadratic variation hM i(t) = t is a BM.
(a) Let f =

pPn

i=1

x2i . Then
f xi =

xi
,
f

fxi ,xi =

x2i
1
+ .
3
f
f

So we have (we write X for X(t) etc)


n
n 
n
n
X
X
X
Bi dBi X Bi2
1
Bi dBi X 2 n
Bi (t)
n1
dX =
+
+
dt =
3+ =
dBi +
dt.
3
X
X
X
X
X X
X(t)
X(t)
i=1
i=1
i=1
i=1
With respect to the filtration generated by B1 , . . . , Bn the process
Bi (t)
dBi
X(t)
is a martingale with continuous paths, hence also the process Y with
dY (t) =

n
X
Bi (t)
i=1

X(t)

dBi

is a martingale with continuous paths. Squaring and using the stated rule for dBi dBj
we get
n
X
Bi2
2
dt = dt.
(dY ) =
X2
i=1

It follows that the quadratic variation is hY i(t) = t. By Levys theorem Y is a BM,

hence X satisfies
n1
dt,
dX(t) = dY (t) +
X(t)
with Y a BM.
2. Let X(t) = B(t) + t be a BM with dift. Use Girsanovs theorem to derive the
joint density of X(t1 ), X(t2 ), X(t3 ) for t1 < t2 < t3 .
Let f (x1 , x2 , x3 ) be the joint density of of X(t1 ), X(t2 ), X(t3 ) when the drift coefficient is .
Suppose first that = 0; then we deal with the BM. We know that X(t3 )
X(t2 ), X(t2 ) X(t1 ), X(t1 ) are independent mean-value normal random variables,
with variances t3 t2 , t2 t1 , t1 . The Jacobian for transition from X(t1 ), X(t2 ), X(t3 )
to X(t1 ), X(t2 ) X(t1 ), X(t3 ) X(t2 ) equals 1, because the transition matrix

1
0 0
1 1 0
0 1 1
has determinant 1. Thus the joint density of X(t1 ), X(t2 ), X(t3 ) (assuming = 0) is
2

2 /(t t )
2
1

f0 (x1 , x2 , x3 ) = cex1 /2 e(x2 x1 )

where
c=

1
(2)3/2

2 /(t t )
3
2

e(x3 x2 )

t1 (t2 t1 )(t3 t2 )

e to P with the Radon-Nikodym derivative dP/dP

e
Changing the measure from P
=
Z, where
2
2
Z = eB(t3 ) t3 /2 = e(X(t3 )t3 ) t3 /2 ,
we achieve that (X(t), t [0, t3 ]) is a BM (Girsanovs theorem). Conversely, to pass
e we need the Radon-Nikodym derivative Z 1 . To obtain f we just need
from P to P
to multiply f0 with Z 1 , the latter considered as a function of x3
f (x1 , x2 , x3 ) = f0 (x1 , x2 , x3 ) exp((x3 t3 ) + 2 t3 /2) =
f0 (x1 , x2 , x3 ) exp(x3 2 t3 /2)
Re-combining the exponents
(x2 x1 )2 (x3 x2 )2
x21

+ x3 2 t3 /2 =
2t1
2(t2 t1 )
2(t3 t3 )
(x1 t1 )2 (x2 x1 (t2 t1 ))2 (x3 x2 (t3 t2 ))2

2t1
2(t2 t1 )
2(t3 t2 )

we obtain
f (x1 , x2 , x3 ) =


2
2
(x1 t1 )
(x2 x1 (t2 t1 ))
(x3 x2 (t3 t2 ))2
c exp

2t1
2(t2 t1 )
2(t3 t2 )
2

which can be written as

2
Y



(xj+1 xj (tj+1 tj ))2
c
,
exp
2(t

t
)
j+1
j
j=0
where t0 = x0 = 0.
Comment: the same could be derived more directly by inspecting the increments
of the BM with drift.
3. For which constant , the process S(t) = exp(B(t) + t) is a martingale?
We compute using Ito formula
1
dS = SdB + Sdt + 2 Sdt.
2
The dt terms vanishes if = 2 /2, hence under this condition we get a martingale.
4. Show that the Ornstein-Uhlenbeck process
X(t) = e

x+e

es dB(s)

satisfies the SDE

dX(t) = X(t)dt + dB(t).
Compute the differential

dX = e

x e

es dB(s) + et et dB(t) = X(t)dt + dB(t).

5. Calculate the quadratic variation hXi(t) for X(t) = eB

2 (t)

We have dX = 2BeB dB plus a dt term, so

Z t
Z T
2
2
2B 2 (t)
E[4B 2 (t)e2B (t) ]dt.
hXi(T ) = E
4B (t)e
dt =
0

The expectation is
2

2B 2 (u)

E[4B (u)e

1
]=
2T

2 /(2t)

ex

4x2 e2x dx.

If t > 1/4 the integral is infinite, therefore hXi(T ) = for T 1/4. For T < 1/4
Z T
4
hXi(T ) =
dt
1/2
1
(t 2)3/2
0 t